This thesis describes a new methodology used in conjunction with artificial intelligence tools to create multiple models for prediction of preterm birth in obstetrical environments. The data mining approach integrates: Decision Trees (DTs), Artificial Neural Networks (ANNs) -specifically a Feed Forward Back Propagation ANN, and Case Based Reasoning System (CBRS).This work also introduces a 5by2 cross validation method, assesses two methods of attribute selection, and considers data prevalence (15% and 8.1%) in training and testing networks. Two databases were assessed from two countries: BORN (Canada) and PRAMS (USA). Best BORN results used selection method 2, had sensitivities of 50.53%, 53.96%, specificities of 91.61%, 95.40%, and area under curves (AUC) of 0.7721, 0.7970 for Parous and Nulliparous cases respectively. Best PRAMS results used selection method 1, had sensitivities of 68.15%, 40.35%, specificities of 64.71%, 94.57%, and area under curves (AUC) of 0.8452, 0.7064 for Parous and Nulliparous cases respectively. iii Above all, challenge yourself. You may well surprise yourself at what strengths you have, what you can accomplish. -Cecile Springer iv
AcknowledgementsFirst and foremost, I would like to take this opportunity to thank my thesis supervisor, Dr. Monique Frize, for her support, encouragement and guidance. I am very grateful to have had the opportunity to work with you.